This paper explores the complex drivers of modern slavery using machine learning methods. The authors address challenges of data scarcity and high dimensionality by employing non-linear machine-learning models, strict cross-validation, and novel variable importance techniques, including Rashomon-set analysis. The model, applied to data from 48 countries, highlights the importance of factors like a country's capacity to protect women's physical security and resource access, particularly for women's vulnerability to exploitation. The model also generated out-of-sample estimates for countries lacking survey data.

Rosa Lavelle-Hill, Gavin Smith, Anjali Mazumder, Todd Landman, James Goulding